Run a risk check on an AI-assisted workflow before rollout

Run a risk check on a workflow where we use, or plan to use, AI. The workflow: [STEPS, and where AI is involved] AI tools and how they are accessed: [e.g. company-approved assistant, public chatbot, automation tool] Data involved: [e.g. customer data, employee data, financials, contracts, code] Who sees or acts on the output: [internal team, customers, regulators, public] Our AI policy or rules: [PASTE, or "none yet"] Industry or regulations that apply: [DETAILS] Check for these risk types: - Accuracy - hallucinated facts, wrong numbers, outdated information. - Data privacy and confidentiality - personal or sensitive data in prompts, retention, sharing. - Security - access permissions, prompt injection from emails or documents, connected tools taking actions. - Bias and fairness - outputs that affect people differently. - Compliance and legal - regulations, IP and copyright, disclosure obligations. - Over-reliance - skipped reviews, lost skills, no fallback if the tool fails. - Reputation - tone or errors reaching customers or the public. Return: 1. A risk register table: Risk | Where in the workflow | Likelihood (H/M/L) | Impact (H/M/L) | Existing control | Recommended control | Owner. 2. The 3 risks to address before rollout. 3. Required human review points. 4. What to monitor after launch, and what would trigger pausing the workflow. Don't overstate legal conclusions; mark items to [CONFIRM WITH LEGAL / COMPLIANCE / IT SECURITY].